A Historical Overview of the Status of Function Words in Dependency Grammar

Timothy Osborne Daniel Maxwell Zhejiang University US Chess Center Hangzhou Washington DC China USA [email protected] [email protected]

(1) eaten Abstract kids have school

This contribution provides a historical The at overview of the analysis of function words in a. The kids have eaten at school. surface syntactic dependency hierarchies. Starting with Tesnière (1959), the overview The USD analysis shown with (1a) stands in progresses through some prominent voices in contrast to more conventional analyses, which the history of DG (Mel'čuk 1958, 1963, Hays position auxiliaries as heads over content verbs 1964, Matthews 1981, Schubert 1987, and prepositions as heads over their nouns: Maxwell and Schubert 1989, Hudson 1976, 1984, etc.). The overview establishes that the (1) have analysis of prepositions has been almost kids eaten unanimous: they are positioned as heads over The at their nouns. There has been more variation concerning the status of auxiliary verbs, school although most DG grammarians have viewed them as heads over their content verbs. b. The kids have eaten at school. Concerning determiners, the dominant position In pursuing the analysis in (1a), USD is advocat- is that they are dependents under their nouns, ing an understanding of surface dependencies although there are a couple of prominent that is generally contrary to the views about voices that assume the opposite stance. function words that have crystallized over the decades in support of the analysis in (1b). 1 The dependency status of function This contribution surveys some prominent words voices in the DG tradition, in order to determine the extent to which the analysis in (1a) has been The distinction between function words and con- advocated over the decades since Tesnière (1959) tent words has been made by linguists of various and the period in which computational linguistics backgrounds. A rough definition of the distinction has become influential. Space does not allow us might be that content words can be understood to try to evaluate these analyses in detail, but without any supporting context whereas function when the authors of these analyses attempt to words cannot. The discussion in this article takes justify their decisions, we occasionally make place in terms of specific syntactic categories comments about their argumentation. which happen to be function words according to this definition. 2 Tesnière (1959/2015) A recent proposal for surface analyses of dependencies categorically subordinates function Tesnière’s analysis of function words has not words to content words. Universal Stanford De- survived into most modern DGs. The reason it pendencies (USD) (de Marneffe et al. 2014) ad- has not done so is tied to the fact that Tesnière’s vocate an annotation scheme that positions aux- subtheory of transfer has also not survived into iliary verbs, adpositions, subordinators, and de- most modern DGs. Tesnière viewed most func- terminers as dependents of the content words with tion words as translatives (auxiliary verbs, pre- which they co-occur. Thus according to this positions, subordinators, many determiners). As scheme, the DG analysis of the sentence The kids such, they were not granted autonomy in the have eaten at school would be as follows: syntactic representations, but rather they appeared together with a content word, the two

241 Proceedings of the Third International Conference on Dependency Linguistics (Depling 2015), pages 241–250, Uppsala, Sweden, August 24–26 2015. forming a dissociated nucleus. What this means Each of the top bubbles encloses the words of a in the current context is that Tesniere’s Éléments single nucleus. Neither word in a bubble is head (1959) did not provide much direct guidance over the other. concerning the dependency analysis of function This greater point established, one can never- words. theless examine Tesnière’s analysis of disso- While Tesnière is widely credited as the father ciated nuclei more closely. When one does so, of modern DGs, he himself was not aware of the one sees that he actually did provide some indi- dependency vs. constituency distinction. That rect guidance concerning the dependency analy- distinction was first established a few years after sis of function words. He drew a distinction be- his death. Hays (1964) is generally credited with tween the structural and the semantic function of establishing the term dependency grammar (as a nucleus (Chapter 29). In a disassociated nuc- opposed to phrase structure grammar). Thus the leus, one of the words guarantees the structural fact that Tesnière’s account of function words left integrity of the nucleus, and the other its seman- much room for debate about the actual depen- tic integrity. He also comments (Chapter 29, §18) dency status of function words should not be so that from an etymological point of view, the sub- surprising. Tesnière never intended to produce an sidiary word in a dissociated nucleus was once account of function words that would be consis- dependent on the constitutive word. tent with the purely dependency-based theories Tesnière later (Chapter 38, §19) states that of syntax that followed him. auxiliary verbs, despite the fact that they lack The relevant aspect of Tesnière’s grammar semantic content, are constitutive of the nucleus can be seen with the sentence Bernard est frappé in which they appear. Thus given these state- par Alfred ‘Bernard is hit by Alfred’, the struc- ments, one can extrapolate that, had Tesnière ture of which Tesnière showed with his Stemma been forced to choose, he would have viewed the 95: auxiliary verb as head over the content verb. Us- ing his conventions, he might have produced an (2) est frappé analysis of the sentence Alfred est grand ‘Alfred Bernard par Alfred is big’ along the following lines:

The noteworthy aspect of this stemma is the (4) .est manner in which est frappé and par Alfred together occupy a single node each time. Neither grand can the function word be viewed as head/parent over the content word nor can the content word .Alfred be viewed as head over the function word.1 For The two words est and grand together still form a Tesnière, the two words est and frappé, and the single nucleus, but now the auxiliary verb is two words par and Alfred, formed a single nuc- shown as head over the adjective, and at the leus together each time. same time, the two together form a single head When Tesnière wanted to draw attention to over the noun Alfred. the fact that a given function word forms a nuc- One can also extend this extrapolated analysis leus together with a content word, he put the two to prepositions. Tesnière viewed a preposition as in a bubble. His stemmas of the sentences Alfred subsidiary to the noun with which it forms a dis- est arrivé ‘Alfred has arrived’ (Stemma 27) and sociated nucleus (Chapter 29, §4). Thus for the Alfred est grand ‘Alfred is big’ (Stemma 28) are phrase livre d’Alfred, lit. ‘book of Alfred’, the given here: following analysis reflects the distinction be-

2 tween constitutive and subsidiary words inside (3) est arrivé est grand the nucleus:

a. Alfred b. Alfred (5) livre

Alfred 1 One of us coauthors prefers the term parent to head ..d’ in this context, reserving the latter term for one specific kind of parent, namely those which are content In other words, Tesnière would probably have words rather than function words. 2The use of est ‘is’ in what is often considered two preferred an analysis that views prepositions as different ways is taken up in section 3.2. dependents of their nouns, if forced to choose.

242 Concerning the other two relevant types of formalization of dependency-based rewrite rules, function words, i.e. subordinators and articles, as inspired directly by Chomsky (1957), and on Tesnière viewed subordinators as translatives the extent to which dependency and constituency that were essentially the same as prepositions in formalisms are weakly or strongly equivalent. how they function. Translatives are part of Tes- One can merely glean a sense of Hays’ under- nière’s theory of transfer, developed in the standing of function words from the article. second half of his book. The purpose of a trans- In particular, Hays discusses the ambiguous lative is to transfer a content word of one cate- sentence They are flying planes, and he captures gory to a content word of another category, e.g. a the ambiguity with the following trees: noun to an adjective, an adjective to an adverb, a (6) o verb to noun, etc. Thus he probably would have favored subordinating them to the verb with o o o which they co-occur inside the nucleus. And a. They are flying planes. concerning articles (definite and indefinite), he o took a varied stance, viewing them as translatives . .o o when they perform a translative function, but as mere dependents of their noun when they do not o perform a translative function. b. They are flying planes. 3 Some early works Tree (6a) reflects the meaning ‘They are making the planes fly’, whereas tree (6b) reflects the 3.1 Mel'čuk (1958, 1963, 1974, etc.) meaning ‘They are planes that are flying’. Hays does not discuss the varying status of the aux- Unlike Tesnière, Igor Mel'čuk has been clear and iliary/copula are in these cases. But from the consistent about the dependency status of func- trees the reader can see that are is taken to be an tion words in surface syntax. Mel'čuk and his auxiliary verb in tree (6a) and a copular verb in collaborators have consistently subordinated tree (6b). content verbs to auxiliary verbs, nouns to adposi- Hays’ interpretation of the auxiliary/copula tions, verbs to subordinators, and determiners to be is representative of how it was widely viewed nouns in surface syntax in their prolific depen- at the time. Chomsky (1957: 38f.) viewed aux- dency-based works on syntax and grammar in the iliaries as a separate class (Aux), meaning he did MTT (Meaning to Text Theory) framework and not classify them as verbs. Thus when a form of otherwise. Mel'čuk and his collaborators have be appears with a main verb, it was deemed an been doing this since his earliest works, starting auxiliary, but when it appeared in the absence of in 1958 (Mel'čuk 1958: 252–4, 1963: 492–3, a main verb, it was deemed a copula. For Hays 1974: 221–4). These early works are in Russian, then, are in (6a) was an auxiliary, whereas are in but the approach is consistent with the prominent (6b) was a copular verb. MTT works in English from the 1980s (e.g. From a modern perspective, the distinction Mel'čuk and Pertsov 1987, Mel'čuk 1988). Hays was building on cannot be maintained. Di- Judging by the dates of these early publica- agnostics for identifying auxiliary verbs reveal tions, it seems likely that Mel'čuk’s works earn that the two putative types of be behave the same the honor of being the pathfinder in this area, in important ways: since the majority of DGs that have come later have done the same, as will become increasingly (7) a. Are they flying planes? clear below. Mel'čuk’s position concerning the b. They are not flying planes. status of function words has been and is particu- c. They are flying planes, and they are, too. larly firm. The words flying planes show the same ambigu- 3.2 Hays (1964) ity in all three of these sentences, just as in Hays’ example. Thus the putative distinction between Hays (1964) is considered a milestone in the de- auxiliary be and copular be lacks an empirical velopment of dependency theory, in part because basis, since the two show the same syntactic be- the article appeared in such a prominent journal, havior with respect to subject-auxiliary inversion Language. Hays’ article does not, however, say (7a), sentence negation (7b), and VP-ellipsis (7c). much about the dependency status of function The two are in fact the same type of verb; they words. The article concentrates instead on the are both auxiliary be.

243 3.3 Hudson (1976, 1984, 1990) This basic constituency-based analysis is present in syntax textbooks from the 1970s (e.g. Bach Richard Hudson’s dependency-based framework 1974, Emonds 1976, Baker 1978). Word Grammar (1984, 1990) has consistently Based in part on Pullum and Wilson’s 1977 taken auxiliary verbs as heads over content verbs, article, a stance took hold in the 1980s that prepositions as heads over nouns, and subordi- viewed auxiliaries of every sort (aspect, voice, nators as heads over verbs. And concerning de- and modality) to be syntactically like full verbs. terminers, Hudson has mostly preferred an anal- For example, they distinguish between present ysis that positions determiners as heads over and past tense and agree with the subject. In Eu- nouns (Hudson 1984: 90–2). Thus Hudson’s ap- ropean languages other than English, both full proach concerning function words is consistent verbs and auxiliaries behave the same way, as insofar as function words are heads over the already discussed in section 3.2. For this reason, content words with which they co-occur. auxiliaries should be granted the status of verbs Concerning the latter issue, i.e. the status of in the hierarchy. This led to analyses like the determiners, the determiner phrase (DP) vs. noun following one, which shows the auxiliary as head phrase (NP) debate has been an ongoing dispute over the content verb, e.g. since the 1980s. Interestingly in this regard, Hudson’s position is a minority one in the DG (8) S community in general, but it certainly finds much NP .VP support among generative grammarians, the ma- 1 jority of whom presently pursue a DP analysis of V1 VP nominal groups. This issue is not addressed here. b. They have succeeded. The discussion focuses instead on auxiliary verbs. Analyses along these lines can be found in many By the time of Hudson’s 1976 book, he had textbooks of the era (e.g. Haegeman 1991, Napo- apparently become convinced that auxiliary li 1993, Ouhalla 1994), and despite the addition verbs are heads over content verbs (p. 149–51), of numerous varied functional categories, the and in his 1984 book Word Grammar, Hudson basic hierarchy of verbs shown with (8b) remains writes in this regard: intact in many recent constituency grammars (e.g. Culicover 2009, Carnie 2013). “It is now widely accepted that a main The interesting aspect of this trend in consti- verb is syntactically subordinate to its tuency grammars, i.e. the trend toward auxiliary auxiliary verb (Pullum and Wilson 1977 is verbs as heads over content verbs, is the fact that particularly important collection of evi- Hudson’s dependency-based system (and dence), and I have accepted this analysis Mel'čuk’s) was ahead of its time. On a depen- in all my dependency analyses.” (p. 91) dency-based analysis, there are just two basic The reference here to Pullum and Wilson (1977) possibilities for the hierarchical analysis of aux- is pointing to a significant debate that took place iliaries that must be considered: either the aux- in the 1970s concerning the status of auxiliary iliary is head over the content verb, or vice versa: verbs (Ross 1969, Chomsky 1972, Huddleston (9) have 1974, Pullum and Wilson 1977). The question concerned the extent to which auxiliaries should They understood be viewed as verbs at all. a. They have understood. As mentioned above, Chomsky (1957) took auxiliaries to be a syntactic class that was to a understood large extent distinct from that of content verbs, They have labeling this class simply Aux (p. 39). This view of auxiliaries led to a ternary-branching analysis b. They have understood. of basic clause structure in which an auxiliary is Apparently Hudson had already decided firmly present, as in (8a): in favor of the analysis in (9a) by 1976. In this (8) S regard, the trend in constituency grammars was lagging significantly behind the stance that Hud- . . NP Aux VP son and Mel'čuk had adopted early in the devel- a. They have succeeded. opment of the Word Grammar and the Mean- ing-Text frameworks.

244 3.4 Matthews (1981) ogy: just like determiners determine their head nouns, auxiliaries determine their head verbs. To Matthews (1981) discusses the distinction be- him, this also meant that the hierarchical rela- tween dependency- and constituency-based syn- tionships between determiner and determined tax at length, and in this regard his book Syntax should be the same in both constructions. was a major contribution to our developing un- The second motivation Matthews produced in derstanding of the distinction between depen- favor of auxiliaries as dependents of their content dency- and constituency-based systems. In the verbs was subcategorization. He briefly dis- book, Matthews took content verbs to be heads cussed the example has appeared (p. 63). Ac- over auxiliaries, prepositions to be heads over cording to Matthews, appeared influences the their nouns, and nouns to be heads over their de- syntactic category and semantic content of the terminers. noun with which it appears, whereas has lacks Two examples from the book illustrate Mat- this ability. Matthews wrote: thews’ positions: “As a form of APPEAR it can take just a (10) a. No animal shall wear clothes. subject (He has appeared) but not both a (p. 155) subject and an object (*He has appeared the speech or *He has appeared Cicero). b. leave the meat in the kitchen For other lexemes it can be the reverse: (p. 87) He has distributed the speech or He has visited Cicero, but not *He has distributed The important observations here are that the or *He has visited. A relation is thus es- modal auxiliary shall is dependent on the content tablished between appeared, or the mor- verb wear in (10a), that the noun kitchen is de- pheme APPEAR, and a subject element. pendent on the preposition in in (10b), and that But at that level the relation of appeared the determiners no, the, and the are dependent on to has, or of the morpheme APPEAR to the their nouns. discontinuous HAVE...past participle, is Concerning prepositions as heads over their quite incidental.” (p. 63) nouns, Matthews did not motivate his position To restate Matthews’ point in other words, con- empirically, but rather appealed to traditional tent verbs influence their linguistic environment case government. He wrote “Grammarians also in a way that auxiliary verbs do not, and for this talk of prepositions having objects…, or having reason, auxiliary verbs should be subordinated to complements…” (p. 146). Every dependency content verbs. hierarchy in Matthews’ book containing a prepo- A noteworthy aspect of Matthews’ reasoning sition shows the preposition as head over its in this area concerns its lexico-semantic nature. noun. Matthews overlooked the fact that from a purely Concerning the status of determiners, Mat- syntactic point of view, it is the finite verb (i.e. thews did not produce specific empirical evi- the auxiliary verb) that licenses the appearance of dence in favor of determiners as dependents of the subject, not the nonfinite verb, e.g. He has their nouns, but rather he appealed to the fact that gone home, *He gone home, He goes home; She they form a closed class. His assumption was that has eaten a lot, *She eaten a lot, She eats a lot. closed class categories are more appropriately From this point of view, there should be a direct viewed as dependents than as heads. The position dependency linking the subject to the finite verb. Matthews was taking concerning determiners was not controversial at that time, so there would 3.5 Schubert(1987)/Maxwell & Schubert(1989) have been little reason for him to justify his decision in the area with further empirical observa- Using the dependency relations of Schubert tions. (1987), Maxwell and Schubert (1989) gathered Concerning the status of auxiliary verbs, annotation schemes from a number of authors for however, Matthews had a bit more to say. He machine translation of a number of languages motivated their status as dependents of content (Bangla/Bengali, Danish, English, Esperanto, verbs in two ways. The first was to point to their Finnish, French, German, Hungarian, Japanese, status as a closed class, and as a closed class, Polish). The project, based in the Netherlands, they were like determiners and thus should be was known as Distributed Language Translation viewed as dependents. He was drawing an anal- (DLT, 1984–1990). DG was used to provide

245 syntactic representations of sentences in a source While the two exceptions just noted provide language, in Esperanto (the intermediate lan- insight into the difficult choices that had to be guage), and in a target language. Grammars made by the authors who participated in the written for this project aimed to show the struc- project, the greater point is that there was a large tural relationships to be derived by automatic measure of agreement concerning the status of parsing. most function words. With the exception of de- In the current context, how function words terminers, most function words were taken to be were dealt with in the various languages is the heads of the content words with which they point of interest. In Schubert’s analysis of Espe- co-occur. ranto, auxiliary verbs are heads over content verbs, common nouns are heads over determiners, 4 The German schools and subordinators are heads over verbs. Schubert (1987: 45) states that his understanding of DG DG has enjoyed particular favor in the German was influenced by the Mannheim school of DG speaking world. German grammarians recog- (Engel 1982). nized the potential of dependency-based syntax These patterns were followed for all other early on. This early recognition may have been grammars with two exceptions. In Danish, sub- due to the particular compatibility of dependen- ordinators are positioned as dependents of the cy-based syntax, which emphasizes verb central- following finite verb rather than as the head of ity, with the verb second (V2) principle of Ger- the subordinate clause. Here is Ingrid Schubert’s man (and other Germanic languages). The finite (1989: 58) statement on this matter: verb is anchored in second position in German declarative clauses, thematic material tending to “These clause introducers may under cer- precede this position and rhematic material tend- tain circumstances be omitted in Danish. I ing to follow it. have not decided to let them be governed The interesting and noteworthy point about by a subordinating conjunction, but to the German schools is the unanimity that one consider the verb of the subordinate con- encounters among the leading voices. DG gram- struction a direct dependent of the verb in marians (Kunze 1975, Lobin 1993, Engel 1982, the superordinate sentence.” 1994, among others) are mostly unanimous in the Perhaps these cases are something like the omis- basic hierarchical analyses of function words that sion of the complementizer that in English, they assume: auxiliary verbs are heads over con- which makes no contribution to meaning and tent verbs; prepositions are heads over nouns; accordingly can often be left out, as in Say (that) and subordinators are heads over verbs. The only it’s true. It is arguably the only subordinator area where one encounters some variation among which has this property. If so, it seems wrong to these experts concerns determiners. The majority base the analysis on this one instance. The alter- stance is certainly that nouns are heads over de- native of an empty node could be considered. terminers, but Lobin (1993) takes the opposite The other exception is in Lobin’s (1989) stance, as he does in Lobin (1989), already men- analysis of German. The determiner rather than tioned in section 3.5, and Eroms (2000) has ar- its noun is taken as the head of the nominal gued for interdependence between article (defi- group. Lobin justifies his analysis in this area by nite or indefinite) and noun. pointing to cases like the following one: Due to the large measure of agreement concerning the hierarchical status of most function (11) unsere Fahrt an die See und eure words, the German-language DG world can be our trip to the sea and yours viewed as speaking with a single voice, and this in die Berge voice is particularly loud by virtue of the fact DG in the mountains enjoys a prominence at schools and universities that is not generally encountered outside of the ‘our trip to the sea and yours to the German-speaking world. A point of interest, mountains’ perhaps, is the reasoning that one finds in the The absence of Fahrt in the second part of the German-language DG literature about the sen- coordinate structure forces one to view eure as tence root. In a two-verb combination such as hat the head of the nominal group (or to posit an gelegt ‘has layed’, the German schools of DG empty nominal node). unanimously view hat as head over gelegt. It is worth considering briefly why they do so.

246 Engel (1994:107–109) points to facts about published works. For this reason, the two lin- subcategorization. He sees the lexical stem hab guistic observations produced next are based on of the auxiliary hat determining the form of the personal communication with Jarmila Panevová, nonfinite verb gelegt as a participle. This rea- one of the founding members of the Prague soning does not work in the opposite direction, school of DG. that is, one cannot view the lexical stem leg- as Worth noting first, though, is that the PDT determining the syntactic form of the auxiliary annotation scheme subordinates only non-modal hat. The insight can then be extended to all auxiliary verbs to content verbs. Modal verbs, in manner of auxiliary verbs. For instance in a contrast, are heads over their content verbs. What combination such as wird wollen ‘will want’, the this means is that the PDT annotation scheme for modal auxiliary wird subcategorizes for the infi- surface syntax deviates from the majority posi- nitive form of the stem woll-, but not vice versa, tion only regarding a single auxiliary verb, that is, the lexical stem woll- does not determine namely být ‘be’ (in all its forms). the syntactic subcategory of the auxiliary wird. There are two linguistic motivations for sub- A related issue concerns the motivation for ordinating the forms of být to the content verb. positioning the finite verb as the root of the One of these concerns a general aspect of the clause. Eroms (2000: 129ff.) motivates the hie- verb ‘be’ in Slavic languages in general. Many rarchical status of auxiliary verbs in another way. Slavic languages lack or omit the finite form of He appeals to the fact that when an auxiliary verb this verb in certain environments. Czech omits a and full verb co-occur, it is the auxiliary that is form of this verb in the 3rd person of the com- finite, not the full verb. The auxiliary verb then pound past, but a 1st and 2nd person form of this bears the functional information of person, num- verb appear in such environments: ber, tense, and mood. The nonfinite verb typical- (12) a. Já jsemspal. (masculine) ly does not express this information. Thus in the I am.slept example from the previous paragraph, i.e. hat ‘I slept.’ gelegt ‘has layed’, the finite auxiliary verb hat expresses number (singular), person (3rd person), b. Já jsemspala. (feminine) tense (present), and mood (indicative). The parti- I am.slept ciple form gelegt, in contrast, can be construed as ‘I slept.’ helping to convey perfect aspect only. This func- c. John spal. tional load that the auxiliary verb bears motivates ‘John slept.’ its status as the root of the clause, i.e. as head over the content verb. If the clitic auxiliary jsem- is subordinated to spal/spala, then there is a direct dependency that 5 The Prague school connects the subject to spal in each of these three cases. But if spal/spala is subordinated to jsem-, The Prague school of DG, as associated with the then an asymmetry appears: the subject is at Prague Dependency Treebank (PDT), agrees times (in the 1st and 2nd person) an immediate with most of the other DG mentioned in this dependent of the auxiliary verb, and at other contribution concerning the hierarchical status of times (in the 3rd person), it is an immediate de- adpositions; they are heads over their nouns. pendent of the participle. One thing that is dif- However, the annotation scheme for the PDT ferent about this construction from others dis- (Hajič 1998) began subordinating auxiliary verbs cussed earlier is that here the auxiliary and main to content verbs in 1996. This aspect of the PDT verb form a single word. The question therefore remains anchored today in the analytical level concerns the extent to which this asymmetry (surface syntax) of the PDT. Due to the promi- should influence choices about the syntactic hie- nence of the PDT in the development of DG rarchy. theory in general, the linguistic motivation for its A possible drawback of this choice is that it choice to subordinate auxiliary verbs to content forces the PDT to draw a distinction between verbs in surface syntax is worth considering, auxiliary být and copular být, since when a form however briefly. of být is the only verb present, e.g. Mary je velká There has been a difficulty in this area, ‘Mary is tall’, the PDT positions that verb as though. Attempts to locate the linguistic motiva- head over the predicative expression (here je ‘is’ tion behind this aspect of the PDT annotation is head overvelká ‘tall’). Examples (7a–c) above scheme have not turned up anything concrete in illustrate that there is no syntactic motivation for

247 distinguishing between an auxiliary be and a co- 6 Concluding discussion pular be in English. Another linguistic argument for subordinating This survey has revealed that there is little sup- forms of the auxiliary být to its content verb is port in the sources examined above for the Uni- seen when multiple forms of být appear in one versal Stanford Dependencies’ (USD) decision to and the same clause: categorically subordinate function words to content words. Not one of the sources surveyed (13) John by byl býval spal. clearly supports the USD analysis of adpositions, John be been been slept and only two of the sources provide support for ‘John would have slept.’ the USD decision to subordinate auxiliaries to This sentence is an example of the past condi- content words. The dominant position, which has tional. Three distinct forms of být appear togeth- crystallized through the decades, is that auxilia- er. By subordinating all three of them directly to ries are heads over content verbs and preposi- spal, a relatively flat structure obtains, and a flat tions are heads over their nouns. And concerning structure has the advantage of avoiding projec- determiners, they are more widely viewed as de- tivity violations (cf. Mel’cuk and Pertsov pendents of their nouns – although their status 1987:181 for a discussion of such a violation in has been the focus of more debate. terms of the arcs used in some forms of DG), The survey has turned up three published ar- which would occur, assuming that certain other guments in support of the USD position and two sequences of these verb forms are possible. unpublished arguments that partially support the Gruet-Skrabalova (2012) offers a different USD position. The published ones are Matthews’ account of the auxiliary být in the course of a argument concerning English auxiliaries in sec- Minimalist analysis of some kinds of ellipsis. tion 3.5, and the arguments concerning Danish She shows that there is a certain degree of free- subordinators and German determiners in section dom in the word order, but the only alternative 3.6. The unpublished ones concerned the Czech order shown by her does not produce projectivity auxiliary být in section 5. violations. In (13), by is the 3rd person (singular De Marneffe et al. do give a few indications or plural) form of the conditional mood and the of the supposed linguistic superiority of USD. only one of the three which is finite. The other The choice of having nouns as heads over adpo- two are both past participles, the first in the per- sitions allows parallelism between prepositional fect aspect and the second in the imperfect aspect. phrases and morphological case-marking (p. The first of them is subcategorized for by the 4585) and also between adpositions and adverbi- preceding conditional verb. Gruet-Skrabovala al clauses (p. 4587). However, it ignores the fact does not discuss any sentence in which the two that adpositions assign case to their complement participles co-occur. However, sentences in this nouns, not vice versa. Hence what one achieves article suggest that both past tense forms of this in the way of more parallelism across the struc- verb subcategorize for a participial complement. tures of distinct languages, one pays for with the From this, we judge that the first participle in (13) unorthodox stipulation that adpositions assign subcategorizes for the second, which in turn case up the syntactic hierarchy to their nouns. subcategorizes for the participial form of the The choice of making predicates heads over main verb. Gruet-Skrabovala states that the final auxiliaries allows parallelism between construc- participle of the auxiliary must directly precede tions which in some languages omit the copula the main verb, although the finite conditional and those which do not (p. 4586). This is true, form can follow the main verb, just like in sub- but alternative solutions such as an empty node ordinate clauses in German. should be considered. Also, if there are several The PDT decision to subordinate forms of být linked auxiliaries, as in might have been dream- to the co-occurring content verb constitutes a ing, they must all have dreaming as their head, so narrow exception to the majority position con- the subcategorization relationship between any cerning function words. The PDT is otherwise two consecutive auxiliaries cannot be shown by consistent with the majority position regarding the dependency links (cf. the discussion of sub- the status of modal verbs, copular verbs, adposi- categorization by Engel in section 4). tions, subordinators, and determiners. De Marneffe et al. note that the choices discussed in their article have a negative effect on parsing:

248 “It is now fairly well-known that … de- Andrew Carnie. 2013. Syntax: A Generative pendency parsing numbers are higher if Introduction. Wiley-Blackwell, Malden, MA. you make auxiliary verbs heads … and if Noam Chomsky. 1957. Syntactic Structures. Mouton, you make prepositions the head of prepo- The Hague. sitional phrases… Under the proposed Noam Chomsky. 1972. Some emperical issues in the USD, SD would be making the ‘wrong’ theory of Transformational Grammar. In S. Peters choice in each case.” (p. 4589) (ed.), Goals of Linguistic Thoery, 63–130. Prentice Parsing accuracy is not the most important crite- Hall, Englewood Cliffs, NJ. ria, however, as the following statements con- Peter Culicover. 2009. Natural Language Syntax. cerning the importance of linguistic quality and Oxford University Press, Oxford UK. downstream applications document: Joseph Emonds. 1976. A Transformational Approach “…it seems wrongheaded to choose a to English Syntax: Root, Structure-Preserving, and Local Transformations. Academic Press, New linguistic representation to maximize York. parser performance rather than based on Ulrich Engel. 1982. Syntax der deutschen the linguistic quality of the representation nd and its usefulness for applications that Gegenwarts- sprache, 2 edition. Erich Schmidt, Berlin. build further processing on top of it.” (p. 4589) Ulrich Engel. 1994. Syntax der deutschen Gegenwarts- sprache, 3rd fully revised edition. Erich Schmidt, For this reason, de Marneffe et al. propose trans- Berlin. forming the USD system to provide two other Hans-Werner Eroms. 2000. Syntax der deutschen results, one for parsing and one called en- Sprache. Walter de Gruyter, Berlin. hanced. Thus by de Marneffe et al.’s own admission, Hana Gruet-Skrabalova. 2012. VP-ellipsis and the parsing accuracy tends to be higher if function Czech auxiliary být ( ‘to be‘). Xlinguae, 2012, 5, 3–15. words are heads over content words, and given the analysis and discussion above, the DG tradi- Liliane Haegeman. 1991. Introduction to Government tion agrees to a large extent that linguistic con- and Binding Theory. Blackwell, Oxford, UK. siderations support most function words as heads Jan Hajič. 1998. Building a syntactically annotated over the content words with which they co-occur, corpus: The Prague Dependency Treebank. In Eva contrary to USD’s stance. Hajičcová (ed.), Issues of Valency and Meaning: If USD wants to claim that linguistic consid- Studies in Honor of Jarmila Panevová.106–32. erations support its unorthodox approach to sur- Karolinum, Charles University Press, Prague. face dependencies, it of course has every right to David Hays. 1964. Dependency theory: A formalism do so in the clash of ideas. But the point we hope and some observations. Language 40, 511–25. to have established in this contribution is that the Rodney Huddleston. 1974. Further remarks on the DG tradition does not support this claim. Quite to analysis of auxiliaries as main verbs. Foundations the contrary, the DG tradition has crystallized of Language 11, 215–29. over the decades to a position that contradicts the Richard Hudson. 1976. Arguments for a non- USD approach. Thus if USD wants to maintain Transformational Grammar. University of Chicago its claim to “linguistic quality”, the burden of Press, Chicago. proof rests firmly on its shoulders; it needs to produce the linguistic reasoning that supports its Richard Hudson. 1984. Word Grammar. Basil position in part by discussing and refuting the Blackwell, New York. observations and reasoning that have coalesced Richard Hudson. 1990. An English Word Grammar. over the decades to the opposite position. Basil Blackwell, Oxford. Ray Jackendoff. 1977. X-bar Syntax: A Study of Phrase Structure. The MIT Press, Cambridge, MA. References Jürgen Kunze.1975.Abhängigkeitsgrammatik. Studia Emmon Bach. 1974. Syntactic Theory. Holt, Rinehart Grammatica 12. Akademie Verlag, Berlin. and Winston, Inc., New York. Henning Lobin. 1989. A dependency syntax of Ger- Carl Lee Baker. 1978. Introduction to Generative man. 17–39. In Maxwell and Schubert. Transformational Syntax. Prentice Hall, Englewood Cliffs, NJ.

249 Henning Lobin. 1993. Koordinationssyntaxalsproze- Lucien Tesnière. 2015 (1959). Elements of Structural durales Phänomen. Studien zur deutschen Gram- Syntax. Translated by Timothy Osborne and Syl- matik 46. Gunter NarrVerlag, Tübingen. vain Kahane. John Benjamins, Amsterdam. Marie-Catherine de Marneffe, Timothy Dozat, Natalia Silveira, Katri Havarinen, Filip Ginter, Joakim Ni- vre, Christopher Manning. 2014. Universal Stan- ford Dependencies: A crosslinguistic typology. LREC, 4585–92 Peter Matthews. 1981. Syntax. Cambridge University Press, Cambridge, UK. Dan Maxwell and Klaus Schubert (eds.). 1989. Meta- taxis in Practice: Dependency Syntax for Multilin- gual Machine Translation. Foris Publications, Dordrecht. Igor Melʹčuk. 1958. Mašinnomperevode s vengersko- gojazykanarusskij — On the automated translation of Hungarian to Russian, Problemykibernetiki, v. 1, 222–64. Igor Melʹčuk. 1963. Avtomatičeskijanaliztekstov (namaterialerusskogojazyka) — Automated analysis of texts (on the basis of Russian). In Slavjans- koejazykoznanie — Linguistique slave. Academy of Sciences of the U.S.S.R., Moscow, 477–509. Igor Melʹčuk. 1974. Opytteoriilingvističeskixmodelej “Smysl - Tekst”. Semantika, Sintaksis. — Sketch of a linguistic model of the sort Meaning – Text. Se- mantics, Syntax. Nauka, Moscow. Igor Mel’čuk and Nikolaj Pertsov. 1987. Surface Syn- tax of English: A Formal Model within the Mean- ing-Text Framework. John Benjamins, Amsterdam. Igor Melʹčuk. 1988. Dependency Syntax: Theory and Practice. State University of New York Press, Al- bany. Donna Jo Napoli. 1993. Syntax: Theory and Problems. Oxford University Press, New York. Jamal Ouhalla. 1994. Transformational Grammar: From rules to Principles and Parameters. Edward Arnold, London. Geoffrey Pullum and Deirdre Wilson. 1977. Auto- nomous syntax and the analysis of auxiliaries. Language 53, 4, 741–88. John Robert Ross. 1969. Auxiliaries as main verbs. Studies in Philosophical Linguistics 1, edited by W. Todd, 77–102. Great Expectations Press, Evanston. Ingrid Schubert. 1989. A dependency syntax of Danish. 39-68, in Maxwell and Schubert. Klaus Schubert. 1987. Metataxis: Contrastive dependency syntax for machine translation. Foris Publi- cations, Dordrecht. Klaus Schubert. 1989. A dependency syntax of Espe- ranto. 207–32, in Maxwell and Schubert. Lucien Tesnière. 1959. Éléments de syntaxe structural. Klincksieck, Paris.

250